An integer linear programming approach for identifying instruction-set extensions

This paper presents an Integer Linear Programming (ILP) approach to the instruction-set extension identification problem. An algorithm that iteratively generates and solves a set of ILP problems in order to generate a set of templates is proposed. A selection algorithm that ranks the generated templates based on isomorphism testing and potential evaluation is described. A Trimaran based framework is used to evaluate the quality of the instructions generated by the technique. Speed-up results of up to 7.5 are observed.